Region-based tracking in an image sequence

Abstract

This paper addresses the problem of object tracking in a sequence of monocular images. The use of regions as primitives for tracking enables to directly handle consistent object-level entities. A motion-based segmentation process based on normal flows and first order motion models provide instantaneous measurements. Shape, position and motion of each region present in such segmented images are estimated with a recursive algorithm along the sequence. Occlusion situations can be handled. We have carried out experiments on sequences of real images depicting complex outdoor scenes.